A machine learning-based characterization framework for parametric representation of liquid sloshing

There is a growing interest in developing a parametric representation of liquid sloshing inside a container stemming from its practical applications in modern engineering systems. The resonant excitation due to sloshing, on the other hand, can cause unstable and nonlinear water waves, resulting in chaotic motions and non-Gaussian wave profiles. This paper presents a novel machine learning-based framework for nonlinear liquid sloshing representation learning. The proposed method is a parametric modeling technique that is based on sequential learning and sparse regularization. The dynamics are categorized into two parts: linear evolution and nonlinear forcing. The former advances the dynamical system in time on an embedded manifold, while the latter causes divergent behaviors in temporal evolution, such as bursting and switching. The effectiveness of the proposed framework is demonstrated using an experimental dataset of liquid sloshing in a tank under horizontal excitation with a wide range of frequencies and vertical slat screen settings immersed in the tank perpendicular to the excitation.